AI News
AI news that matters. Updated daily.
No stories match your filters.
ChatGPT Hits 1 Billion Monthly Active Users, Fastest App Ever
1 billion monthly active users. OpenAI CEO Sam Altman confirmed the number on June 3, making ChatGPT the fastest consumer application in history to reach that scale.
One Settings Change Removes Google AI Overviews From Every Search Result
There's a clean workaround to remove Google's AI Overview summaries from your search results without browser extensions or appending -ai to every query: set your Google search region to France.
Qwen3.5-9B Beats Google's Gemma-4-12B on 5 of 8 Standard Benchmarks
Bigger isn't always better. Community benchmarks comparing Alibaba's Qwen3.5-9B against Google's Gemma-4-12B-it found the smaller model winning on 5 of 8 standard tests - despite having 3 billion fewer parameters (the internal weights that give language models their capability).
Google's Gemma 4 12B Runs Multimodal AI on 16GB RAM, Apache 2.0 License
A 12-billion-parameter AI model that runs entirely on your laptop, handles both text and images, and costs nothing to use. Google released Gemma 4 12B this week under an Apache 2.0 license - no cloud connection, no API calls, no monthly bill.
Google's Gemma 4 Lineup Is Expanding, Possibly With a 120B Model
More Gemma 4 models are on the way. Google has signaled that the family will expand beyond the current 12B instruction-tuned model, with reports pointing toward a 120B parameter variant as one possibility.
Google's Gemma 4 12B Drops the Separate Vision Encoder for a Unified Architecture
Most open-source vision models work by attaching a separate image encoder - a component that converts pictures into tokens a language model can process - onto an existing text model. Gemma 4 12B skips that split. Google's latest open-weight release processes both text and images through a single unified network, without any separate encoder in the pipeline.
NeurIPS 2026 Used an AI Detector to Reject Papers Without Validating It First
69%. That's the AI-generated probability score Pangram reportedly assigned to at least one paper that NeurIPS 2026 desk-rejected for alleged AI-writing policy violations - and according to researcher S. Berezin in a LinkedIn post, the detector was never validated against the actual style of academic machine learning writing before being used to make those decisions.
Google DeepMind Releases Gemma 4 12B Open Model on Hugging Face
Google DeepMind released Gemma 4 12B on Hugging Face on June 3, adding a 12-billion parameter model to its open-weights model family. The model is available for download and local deployment without requiring an API subscription.
Uber's $1,500/Month AI Cap Is a Useful Benchmark for Everyone Else
$1,500 a month. That's the per-employee AI spending cap Uber has reportedly set, a figure that developer and blogger Simon Willison analyzed as one of the more useful public calibration points on what enterprise AI use actually costs.
Opus 4.8 Burned 12 Hours With Zero Output. Sonnet 4.6 Finished the Same Job in One Session.
One model produced zero deliverables after 12 hours. The other finished the same type of project in a single session. Both are from Anthropic, both are named Claude, and the gap between them raises a genuinely useful question about how to choose the right tool for production work.
Perplexity Billing Complaints Mount as Subscribers Hit AI-Only Support Walls
Perplexity Pro subscribers are reporting a frustrating loop: they believe they've canceled, then get charged again, and when they contact support they get routed to an automated AI assistant that won't escalate to a human or process refunds.
OpenAI Proposes Federal Framework for Frontier AI Safety and National Security
Three months into an active congressional push on AI legislation, OpenAI has put its preferred regulatory framework on the table. The company's blueprint, published June 3rd, proposes a federal approach built around three pillars: safety requirements for frontier AI developers before deployment, infrastructure resilience standards, and national security controls on the most capable models.
Solo Founder Hits 1.5M Search Impressions in 3 Months Using Claude as His Entire SEO Team
1.5 million search impressions. Three months. One person.
AI Delivers 7.8% Productivity Gains, Not 10x - And Most Fade Within a Quarter
7.8%. That's how much AI lifted productivity across hundreds of engineers in one operator's measurement - tracked across three companies over real work quarters. Not a vendor study. Not a conference keynote. What actually showed up in the data when someone sat down and measured it.
Microsoft Releases Aion 1.0 Instruct and Plan Models on Azure
Microsoft released two new AI models today: Aion 1.0 Instruct and Aion 1.0 Plan, both available through Azure AI Foundry.
Claude Code Connected to Full Polymarket Trade History via MCP Database
What happens when you give an AI coding assistant full read access to every trade and wallet on Polymarket - a prediction market where people bet real money on real-world events?
Nous Research Launches Hermes Desktop App for Local AI
Nous Research has been releasing fine-tuned open-source models for years. The Hermes series became one of the most-downloaded instruction-tuned model families in the local AI community - built on top of Meta's Llama, Mistral, and similar base models, then further trained ("fine-tuned," in technical terms) to follow instructions more precisely and handle complex tasks without constant hand-holding. The downside was always setup: using Hermes meant configuring Ollama, LM Studio, or a similar tool before you could actually chat with it.
Anthropic Grows Project Glasswing to 150 Enterprise Partners
Project Glasswing, Anthropic's enterprise partner program, has expanded to 150 member companies - a milestone that reflects how AI labs are building the indirect sales channels that large enterprises actually use to buy software.
What Anthropic's Red Team Found Analyzing 832 Malicious AI Accounts
832 accounts. That's how many Anthropic banned for malicious cyber activity between March 2025 and March 2026 - and the company's Frontier Red Team spent that year mapping what those attackers actually did with AI, publishing the full findings this week.
Anthropic Adds Services Track and Partner Hub to Claude Network
Enterprise AI adoption tends to stall not on the technology but on the implementation. Anthropic's latest announcement addresses that problem directly: a Services Track and Partner Hub added to the Claude Partner Network.